Rather than adding hardware, couldn't the existing computers be programmed to compare the aircraft's actual acceleration to that predicted? Surely the existing sensors are accurate enough to permit detection of a major weight error early in the takeoff roll.
But as SLF I have no idea whether this would be solving a real problem, or whether additional automation would just exacerbate the underling problem (if there is one).